Search Space Decomposition Research Articles

Adaptive search space decomposition method for pre- and post-buckling analyses of space truss structures

The paper proposes a novel adaptive search space decomposition method and a novel gradient-free optimization-based formulation for the pre- and post-buckling analyses of space truss structures. Space trusses are often employed in structural engineering to build large steel constructions, such as bridges and domes, whose structural response is characterized by large displacements. Therefore, these structures are vulnerable to progressive collapses due to local or global buckling effects, leading to sudden failures. The method proposed in this paper allows the analysis of the load-equilibrium path of truss structures to permanent and variable loading, including stable and unstable equilibrium stages and explicitly considering geometric nonlinearities. The goal of this work is to determine these equilibrium stages via optimization of the Lagrangian kinematic parameters of the system, determining the global equilibrium. However, this optimization problem is non-trivial due to the undefined parameter domain and the sensitivity and interaction among the Lagrangian parameters. Therefore, we propose to formulate this problem as a nonlinear, multimodal, unconstrained, continuous optimization problem and develop a novel adaptive search space decomposition method, which progressively and adaptively re-defines the search domain (hypersphere) to evaluate the equilibrium of the system using a gradient-free optimization algorithm. We tackle three benchmark problems and evaluate a medium-sized test representing a real structural problem in this paper. The results are compared to those available in the literature regarding displacement–load curves and deformed configurations. The accuracy and robustness of the adopted methodology show a high potential for gradient-free algorithms to analyze space truss structures.

Two-stage hybrid learning-based multi-objective evolutionary algorithm based on objective space decomposition

Multi-objective evolutionary algorithm (MOEA) based on search space decomposition (MOEA/D-M2M) is a promising framework for solving multi-objective optimization problems (MOPs). It is crucial yet challenging for an MOEA to balance convergence and diversity. Few studies have been attempted to deal with the balancing problem under the MOEA/D-M2M framework. In this paper, a two-stage hybrid learning-based MOEA, dubbed HLMEA, is proposed to address this problem. In the first stage, we propose a genetic operator with an adaptive scaling parameter in order to accelerate the convergence of the search, while the environmental selection method maintains the diversity. In the second stage, the K-means clustering method is employed within each subpopulation to construct a mating pool for each individual. In each mating pool, an adaptive method based on hypervolume is proposed to choose a suitable differential evolution operator for offspring generation. Different from the first stage, the environmental selection method used in NSGA-II is applied to select a new population. Thereafter, subpopulations are obtained by dividing the newly created population. Extensive experimental results show that HLMEA has better performance than nine state-of-the-art MOEAs. The ablation study demonstrates the effectiveness of each component of HLMEA.

VolcanoML: speeding up end-to-end AutoML via scalable search space decomposition

End-to-end AutoML has attracted intensive interests from both academia and industry which automatically searches for ML pipelines in a space induced by feature engineering, algorithm/model selection, and hyper-parameter tuning. Existing AutoML systems, however, suffer from scalability issues when applying to application domains with large, high-dimensional search spaces. We present VolcanoML, a scalable and extensible framework that facilitates systematic exploration of large AutoML search spaces. VolcanoML introduces and implements basic building blocks that decompose a large search space into smaller ones, and allows users to utilize these building blocks to compose an execution plan for the AutoML problem at hand. VolcanoML further supports a Volcano-style execution model -- akin to the one supported by modern database systems -- to execute the plan constructed. Our evaluation demonstrates that, not only does VolcanoML raise the level of expressiveness for search space decomposition in AutoML, it also leads to actual findings of decomposition strategies that are significantly more efficient than the ones employed by state-of-the-art AutoML systems such as auto-sklearn. This paper is the extended version of the initial VolcanoML paper appeared in VLDB 2021.

Uninterrupted path planning system for Multi-USV sampling mission in a cluttered ocean environment

This paper presents an uninterrupted collision-free path planning system that facilitates the operational performance of multiple unmanned surface vehicles (USVs) in an ocean sampling mission. The proposed uninterrupted path planning system is developed based on the integration of a novel B-Spline data frame and particle swarm optimization (PSO)-based solver engine. The new B-spline data framing structure provides smart sampling of the candidate spots without needing full stop for completing the sampling tasks. This enables the USVs to encircle the area smoothly while simultaneously correcting the heading angle toward the next spot and preventing sharp changes in the vehicle's heading. Then, the optimization engine generates optimal, smooth, and constraint-aware path curves for multiple USVs to conduct the sampling mission from start point to the rendezvous point. The path generated incorporates controllability over the vehicles' velocity profile to prevent experiencing zero velocity and frequent stop/start switching of the controller. To achieve faster convergence of the optimization routine, a suitable search space decomposition scheme is proposed. Extensive simulation studies emulating a realistic ocean sampling mission are conducted to examine the feasibility and effectiveness of the proposed path planning system. This encapsulates modelling a realistic maritime environment of Indonesian Archipelago in Banda Sea including ocean waves, obstacles, and no-fly zones and introducing several performance indices to benchmark the path planning system performance. This process is accompanied by a comparative study of the proposed path planning system with a well-known state-of-the art piecewise, rapidly exploring random tree (RRT), and differential evolution-based path planning algorithms. The results of the simulation confirm the suitability and robustness of the proposed path planning system for the uninterrupted ocean sampling missions.

Search space decomposition for resource-constrained project scheduling

The mechanism with search space decomposition has been ever proposed for resource-constrained project scheduling which can reduce the search space. Theoretical equivalence for search spaces before and after decomposition is proven in this paper. It is shown that the mechanism can be easily incorporated in diverse algorithms. Four experiential strategies and a theoretical strategy are presented. The decomposition-based approach is nested in three classic algorithms to verify the effectiveness of the mechanism in numerical experiments. The results show that the decomposition-based approach can improve the quality of schedules under certain conditions, and the influence is diverse from the algorithms in which they are incorporated. Moreover, the most suitable decomposition strategies for each algorithm are different. Future work, such as priority-rules based on the decomposition strategies, strategies according to the parameters of activity network, and parallel computing with the decomposition mechanism, may bring new opportunities in project scheduling.

VolcanoML

End-to-end AutoML has attracted intensive interests from both academia and industry, which automatically searches for ML pipelines in a space induced by feature engineering, algorithm/model selection, and hyper-parameter tuning. Existing AutoML systems, however, suffer from scalability issues when applying to application domains with large, high-dimensional search spaces. We present VOLCANOML, a scalable and extensible framework that facilitates systematic exploration of large AutoML search spaces. VOLCANOML introduces and implements basic building blocks that decompose a large search space into smaller ones, and allows users to utilize these building blocks to compose an execution plan for the AutoML problem at hand. VOLCANOML further supports a Volcano-style execution model - akin to the one supported by modern database systems - to execute the plan constructed. Our evaluation demonstrates that, not only does VOLCANOML raise the level of expressiveness for search space decomposition in AutoML, it also leads to actual findings of decomposition strategies that are significantly more efficient than the ones employed by state-of-the-art AutoML systems such as auto-sklearn.

An adaptive differential evolution with decomposition for photovoltaic parameter extraction.

Photovoltaic (PV) parameter extraction plays a key role in establishing accurate and reliable PV models based on the manufacturer's current-voltage data. Owning to the characteristics such as implicit and nonlinear of the PV model, it remains a challenging and research-meaningful task in PV system optimization. Despite there are many methods that have been developed to solve this problem, they are often consuming a great deal of computing resources for more satisfactory results. To reduce computing resources, in this paper, an advanced differential evolution with search space decomposition is developed to effectively extract the unknown parameters of PV models. In proposed approach, a recently proposed advanced differential evolution algorithm is used as a solver. In addition, a search space decomposition technique is introduced to reduce the dimension of the problem, thereby reducing the complexity of the problem. Three different PV cell models are selected for verifying the performance of proposed approach. The experimental result is firstly compared with some representative differential evolution algorithms that do not use search space decomposition technique, which demonstrates the effectiveness of the search space decomposition. Moreover, the comparison results with some reported well-established parameter extraction methods suggest that the proposed approach not only obtains accurate and reliable parameters, but also uses the least computational resources.

A decomposed branch-and-price procedure for integrating demand planning in personnel staffing problems

The personnel staffing problem calculates the required workforce size and is determined by constructing a baseline personnel roster that assigns personnel members to duties in order to cover certain staffing requirements. In this research, we incorporate the planning of the duty demand in the staff scheduling problem in order to lower the staffing costs. More specifically, the demand originates from a project scheduling problem with discrete time/resource trade-offs, which embodies additional flexibility as activities can be executed in different modes. In order to tackle this integrated problem, we propose a decomposed branch-and-price procedure. A tight lower and upper bound are calculated using a problem formulation that models the project scheduling constraints and the time-related resource scheduling constraints implicitly in the decision variables. Based upon these bounds, the strategic problem is decomposed into multiple tactical subproblems with a fixed workforce size and an optimal solution is searched for each subproblem via branch-and-price. Fixing the workforce size in a subproblem facilitates the definition of resource capacity cuts, which limit the set of eligible project schedules, decreasing the size of the branching tree. In addition, in order to find the optimal integer solution, we propose a specific search strategy based upon the lower bound and dedicated rules to branch upon the workload generated by a project schedule. The computational results show that applying the proposed search space decomposition and the inclusion of resource capacity cuts lead to a well-performing procedure outperforming different other heuristic and exact methodologies.

Towards efficient security policy lookup on many-core network processing platforms

Modern network security devices employ packet classification and pattern matching algorithms to inspect packets. Due to the complexity and heterogeneity of different search data structures, it is difficult for existing algorithms to leverage modern hardware platforms to achieve high performance. This paper presents a Structural Compression (SC) method that optimizes the data structures of both algorithms. It reviews both algorithms under the model of search space decomposition, and homogenizes their search data structures. This approach not only guarantees deterministic lookup speed but also optimizes the data structure for efficient implementation on many-core platforms. The performance evaluation reveals that the homogeneous data structure achieves 10Gbps line-rate 64byte packet classification throughput and multi-Gbps deep inspection speed.

GPU Accelerated Multi-agent Path Planning Based on Grid Space Decomposition

In this work, we describe a simple and powerful method to implement real-time multi-agent path-finding on Graphics Processor Units (GPUs). The technique aims to find potential paths for many thousands of agents, using the A* algorithm and an input grid map partitioned into blocks. We propose an implementation for the GPU that uses a search space decomposition approach to break down the forward search A* algorithm into parallel independently forward sub-searches. We show that this approach fits well with the programming model of GPUs, enabling planning for many thousands of agents in parallel in real-time applications such as computer games and robotics. The paper describes this implementation using the Compute Unified Device Architecture programming environment, and demonstrates its advantages in GPU performance compared to GPU implementation of Real-Time Adaptive A*.

Genetic algorithm with automatic termination and search space rotation

In the last two decades, numerous evolutionary algorithms (EAs) have been developed for solving optimization problems. However, only a few works have focused on the question of the termination criteria. Indeed, EAs still need termination criteria prespecified by the user. In this paper, we develop a genetic algorithm (GA) with automatic termination and acceleration elements which allow the search to end without resort to predefined conditions. We call this algorithm “Genetic Algorithm with Automatic Termination and Search Space Rotation”, abbreviated as GATR. This algorithm utilizes the so-called “Gene Matrix” (GM) to equip the search process with a self-check in order to judge how much exploration has been performed, while maintaining the population diversity. The algorithm also implements a mutation operator called “mutagenesis” to achieve more efficient and faster exploration and exploitation processes. Moreover, GATR fully exploits the structure of the GM by calling a novel search space decomposition mechanism combined with a search space rotation procedure. As a result, the search operates strictly within two-dimensional subspaces irrespective of the dimension of the original problem. The computational experiments and comparisons with some state-of-the-art EAs demonstrate the effectiveness of the automatic termination criteria and the space decomposition mechanism of GATR.

Pickup and Delivery問題における探索空間の分割に基づく擬似焼鈍し法の構成

Pickup and Delivery問題における探索空間の分割に基づく擬似焼鈍し法の構成